Page 110 - 臺大管理論叢第32卷第1期

Internet Celebrity Economy: Exploring the Value of Viewers’ Comment Features and Live Streamers’
Marketing Strategies in Forecasting Revenue

streamers may have a stronger impact on viewers’ perceptions. Wan et al. (2017) state that
streamers’ personal characteristics, such as personality and sociability, indirectly influence
donation intentions. From a hosting style perspective, Vraga, Edgerly, Bode, Carr, Bard,
Johnson, Kim, and Shah (2012) indicate that sociability increases perceptions of infor-

mation value and enhances host credibility and that humor influences audience likeability
toward the host. Chang, Zhu, Wang, and Li (2018) further suggest that an advocate’s per-
suasiveness positively influences participants’ attitude changes. These findings imply that

hosting styles, such as sociable, comical, and persuasive, play an important role in influ-
encing viewers’ behaviors. In addition, although there has been no research to examine the
effects of food streamers’ behavior on their revenue, we often see successful food stream-
ers share food features, cooking skills, or tasting experiences during live streaming. Based
on the above discussions, this study infers that streamers’ behaviors (i.e., chatting with the

viewers, responding to viewers’ questions, sharing food features, sharing cooking skills,
and sharing tasting experiences) and characteristics (i.e., gender, physical attractiveness,
and hosting styles) may influence their revenues.

3. Data and Methods

This study examines the effects of viewers' comments features and streamers'
behvaior on viewers' gift-sending behavior during live streaming. To do so, we employ

the following three steps. (1) Data collection and preprocessing: in this step, we use
DouYu as our data source to collect the content of viewers' comments and viewers' gift-
sending numbers in food live streaming. We then use word segmentation methods and
word frequency statistics for data preprocessing. (2) Sentiment analysis: in this step, we

use sentiment analysis to analyze the viewers' comment content. (3) Streamers' behavior
coding: we code streamers' behavior during live streaming.

3.1 Data Collection and Preprocessing
The dataset used to evaluate the gift sending behavior of the proposed models

contained 10 food streamers on the live-streaming platform released from March 1
to April 15, 2019, with comments and gifts sent from viewers, as well as streamers'
characteristics and behaviors. The data consist of 38,183 seconds of live-streaming data,

102